Table 3.
Results for fully connected layer of convolutional neural network by SMAC_NEURON architecture for 468-128-10 structure using approximate multipliers and adders
| Multiplier type | Approximation level | Area | Delay | Latency | Power | Energy | HMR | Area energy (%) | Energy gain (%) | |||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hidden | Output | |||||||||||
| Mul | Add | Mul | Add | |||||||||
| Exact | 0 | 0 | 0 | 0 | 134,651 | 5.77 | 0.0034 | 14.53 | 0.050 | 8.81 | 0 | 0 |
| LEBZAM | 3 | 0 | 0 | 0 | 125,350 | 6.22 | 0.0037 | 11.70 | 0.043 | 10.00 | 7 | 13 |
| LEBZAM | 4 | 0 | 0 | 0 | 122,349 | 5.93 | 0.0035 | 11.88 | 0.042 | 12.21 | 9 | 18 |
| LEBZAM | 3 | 5 | 3 | 5 | 121,482 | 5.80 | 0.0035 | 9.25 | 0.032 | 9.10 | 10 | 36 |
| LEBZAM | 4 | 6 | 4 | 6 | 117,278 | 6.06 | 0.0036 | 8.07 | 0.029 | 10.24 | 13 | 42 |